🚀 unsup-simcse-ja-large
This is a model for feature extraction and sentence similarity tasks, leveraging sentence-transformers and based on the Transformer architecture.
🚀 Quick Start
Prerequisites
Before using the model, you need to install the necessary libraries.
📦 Installation
If you have sentence-transformers installed, using this model becomes straightforward:
pip install -U fugashi[unidic-lite] sentence-transformers
💻 Usage Examples
Basic Usage
from sentence_transformers import SentenceTransformer
sentences = ["こんにちは、世界!", "文埋め込み最高!文埋め込み最高と叫びなさい", "極度乾燥しなさい"]
model = SentenceTransformer("cl-nagoya/unsup-simcse-ja-large")
embeddings = model.encode(sentences)
print(embeddings)
Advanced Usage
Without sentence-transformers, you can use the model as follows: First, pass your input through the transformer model, then apply the appropriate pooling operation on top of the contextualized word embeddings.
from transformers import AutoTokenizer, AutoModel
import torch
def cls_pooling(model_output, attention_mask):
return model_output[0][:,0]
sentences = ['This is an example sentence', 'Each sentence is converted']
tokenizer = AutoTokenizer.from_pretrained("cl-nagoya/unsup-simcse-ja-large")
model = AutoModel.from_pretrained("cl-nagoya/unsup-simcse-ja-large")
encoded_input = tokenizer(sentences, padding=True, truncation=True, return_tensors='pt')
with torch.no_grad():
model_output = model(**encoded_input)
sentence_embeddings = cls_pooling(model_output, encoded_input['attention_mask'])
print("Sentence embeddings:")
print(sentence_embeddings)
🔧 Technical Details
Full Model Architecture
SentenceTransformer(
(0): Transformer({'max_seq_length': 512, 'do_lower_case': False}) with Transformer model: BertModel
(1): Pooling({'word_embedding_dimension': 768, 'pooling_mode_cls_token': True, 'pooling_mode_mean_tokens': False, 'pooling_mode_max_tokens': False, 'pooling_mode_mean_sqrt_len_tokens': False})
)
Model Summary
Property |
Details |
Fine-tuning Method |
Unsupervised SimCSE |
Base Model |
cl-tohoku/bert-large-japanese-v2 |
Training Dataset |
Wiki40B |
Pooling Strategy |
cls (with an extra MLP layer only during training) |
Hidden Size |
1024 |
Learning Rate |
3e-5 |
Batch Size |
64 |
Temperature |
0.05 |
Max Sequence Length |
64 |
Number of Training Examples |
2^20 |
Validation Interval (steps) |
2^6 |
Warmup Ratio |
0.1 |
Dtype |
BFloat16 |
See the GitHub repository for a detailed experimental setup.
📄 License
This model is released under the cc-by-sa-4.0 license.
Citing & Authors
@misc{
hayato-tsukagoshi-2023-simple-simcse-ja,
author = {Hayato Tsukagoshi},
title = {Japanese Simple-SimCSE},
year = {2023},
publisher = {GitHub},
journal = {GitHub repository},
howpublished = {\url{https://github.com/hppRC/simple-simcse-ja}}
}